In courtiol/LM2GLMM: Data science for biologists: Generalized linear modelling with R

library(LM2GLMM)
knitr::opts_chunk$set(cache = FALSE,
                      fig.align = "center",
                      fig.width = 4,
                      fig.height = 4,
                      cache.path = "./cache_knitr/Exo_LM_intro/",
                      fig.path = "./fig_knitr/Exo_LM_intro/")
options(width = 110)
set.seed(1L)

Design matrix

Alternative representations of design matrices

If the average IQ for boys were 90 and 110 for girls, what would the parameter values be, for each of the following alternative coding for a boy and a girl (using treatment contrasts):

• "boy" vs "girl"
• "male" vs "female"
• 0 vs 1
• 1 vs 2
• TRUE vs FALSE

[Back to main menu](./Title.html#2)

Alternative representations of design matrices

Are these different representations of ages equivalent?

• "baby" vs "child" vs "adult"
• 0 vs 1 vs 2
• 1 vs 2 vs 3

Simulating data

Turning script into function

Turn the following code into a function that takes the number of aliens to generate as an argument:

Tip: Remember to check ?sample() to understand what this new function does.

Alien <- data.frame(humans_eaten = sample(1:3, size = 6, replace = TRUE), intercept = 50, slope = 1.5)
Alien$error <- rnorm(n = 6, mean = 0, sd = sqrt(5))
Alien$size <- Alien$intercept + Alien$slope*Alien$humans_eaten + Alien$error

Simulating interactions

Adapt the following code to include an interaction between sex and bh_trips.

set.seed(123)
Alien2 <- data.frame(planet = factor(c(rep("Chambr", 2), rep("Riplyx", 2), rep("Wickor", 2))),
                     sex = factor(rep(c("Z", "ZZ"), times = 3)),
                     bh_trips = 6:1)
Alien2$age <- rep(0, nrow(Alien2))
Alien2$age[Alien2$planet == "Chambr"] <- Alien2$age[Alien2$planet == "Chambr"] + 100 
Alien2$age[Alien2$planet == "Riplyx"] <- Alien2$age[Alien2$planet == "Riplyx"] + 1000
Alien2$age[Alien2$planet == "Wickor"] <- Alien2$age[Alien2$planet == "Wickor"] + 10000
Alien2$age[Alien2$sex == "Z"] <- Alien2$age[Alien2$sex == "Z"] + -20
Alien2$age[Alien2$sex == "ZZ"] <- Alien2$age[Alien2$sex == "ZZ"] + 20
Alien2$age <- Alien2$age + 0.5*Alien2$bh_trips
Alien2$age <- Alien2$age + rnorm(nrow(Alien2), mean = 0, sd = 4)
Alien2

Model parameter estimates

The UK dataset

Start by fitting this model:

fit_UK1 <- lm(height ~ drink + sex*weight, data = UK)

Note: for information on this dataset, check the details section in ?UK.

Predictions with fit_UK1

Compute the following predictions by (1) creating a design matrix by hand & (2) using the function predict():

• height of a 30 Kg boy whose mum drank most days?
• height of a 30 Kg girl whose mum drank most days?
• height of a 35 Kg boy whose mum drank 2 to 3 times a week?

newX <- matrix(
  c(1, 1, 0, 0, 0, 30, 0,
    1, 1, 0, 0, 1, 30, 30,
    1, 0, 0, 0, 0, 35, 0),
  nrow = 3, byrow = TRUE)
colnames(newX) <- names(coef(fit_UK1))
newX %*% coef(fit_UK1)

newdata1 <- data.frame(
  drink = c("Most days", "Most days", "2 to 3 times a week"),
  sex = c("Boy", "Girl", "Boy"),
  weight = c(30, 30, 35))
predict(fit_UK1, newdata = newdata1)

Model parameter estimates of fit_UK1

data.frame(coef(fit_UK1))

What do each of these estimates really mean?

(If you struggle interpreting parameter estimates, look at the fitted values)

More predictions with fit_UK1

Create a plot with predictions showing the influence of the weight of girls on their height!

Tip: Look back at slides about predictions to work out how to deal with the effect of 'drink'.

[Back to main menu](./Title.html#2)

courtiol/LM2GLMM documentation built on July 3, 2022, 7:42 a.m.